Finally completed NTU EiMBA's cross-cohort vibe coding closed-door sharing today!
Since learning to deploy in March, our company has directly built systems for finance, project management, and contracts. Most valuably, everyone can be a builder—we can share projects and build AI thinking together.
Since I've benefited tremendously, I've been actively sharing externally. Every time I help someone successfully deploy, I feel incredibly happy! Today was another success!
Why I Decided to Learn Vibe Coding to Deployment
Four years ago, I outsourced to an engineer for the first time. With zero experience, I didn't know how to properly specify requirements or judge if time estimates were reasonable. Every time I wanted to optimize a feature (like email notifications), it cost at least 100,000 NT$. I realized I lacked this capability, and operational costs were too high, so I shut down the brand altogether.
Now vibe coding has finally arrived! But before March, I hadn't learned how to deploy. After handing frontend and backend code to an engineer, it still cost 100,000 NT$. I was heartbroken, and the website dragged on for a long time without launching. That's when I decided to master deployment. In a short time, I created multiple company systems. This post will first introduce case studies, then share my self-learning journey.
Company Background
- The company operates in public relations and entertainment industries.
- Within one month, the team developed and launched multiple internal systems: bonus payout system, project management system, official website, employee training platform, contract AI tool, quotation system, and more.
- Systems are now in use and have replaced some SaaS solutions, saving hundreds of thousands in costs.
Team Collaboration Model and Infrastructure
- AI Thinking: All team members are required to learn AI thinking and vibe coding, supporting cross-device and cross-timezone collaborative development and modifications.
- Unified Account System: Fully use Google Workspace; all AI tools (Gemini, GPT, Claude, Vercel, Canva, etc.) and internal systems log in with the same corporate Google account, achieving knowledge sharing and unified permissions.
- Deployment and Co-editing: Deploy via Vercel; code can be accessed, modified, and co-edited by different people, forming a decentralized website maintenance model.
- Knowledge Preservation: Emphasize completely documenting and preserving the development process and knowledge.
Case Studies
Case Study 1: Finance and Bonus Payout System
- Contract Management: After uploading a contract, AI automatically identifies dates, parties involved, payment dates, and payment installments.
- Project Management: Contract information automatically flows into the project management area to track invoicing and payment progress.
- Expense Audit: Colleagues can audit transportation, insurance, meal expenses, etc. in the system.
- Cost and Profit Calculation: Expenses are matched to corresponding projects and automatically calculate operational costs, gross margins, and generate P&L statements and cash flow statements.
- Transfer and Invoice Reminders: Employees submit transfer requests and reminders to finance to issue invoices; the system can generate PDFs as vouchers.
Special feature: An "eye" function can one-click pixelate sensitive financial numbers (like bank balance).
Case Study 2: OBAsana (Internal Project Management System)
Replaces Asana to save costs (originally ~70,000–80,000 NT$ per year). Core functions mimic Asana, providing daily work areas, team project areas, and supporting multiple views including list, kanban, Gantt chart, calendar, and dashboard.
Case Study 3: Official Website
Migrated from WordPress to a self-developed website. Service items, strategy descriptions, etc. are automatically generated by AI analyzing company Google Drive project data, rather than manually written.
Article Publishing Automation:
- Organize articles and images into a single folder
- Use AI commands to have the system extract folder contents
- AI automatically generates drafts, adds images, and publishes to the frontend
Case Study 4: Corporate Internal Learning Network
Solves repetitive onboarding explanations and reduces training costs. AI deeply mines company knowledge data stored in Google Drive and organizes it into a structured training website. New employees learn along planned paths; the site includes AI-generated quizzes with randomized questions to avoid rote memorization.
Case Study 5: Contract and Quotation Generation System
Provides a web interface where sales staff can select and fill in customer requirements. After entering customer name and requirement options, the system automatically calculates the quote; clicking "Generate Contract" produces a complete contract PDF in the company's standard format.
As a liberal arts student with zero coding experience and no engineers on staff, being able to collaborate this way is truly magical!
Tools Used
- Mac series
- Claude Code terminal
- Vercel deployment
Learning Timeline
| Time | Milestone |
|---|---|
| 2025/10 | Learned Google AI Studio |
| 2026/2/21 | Studied Claude + Open Claw during Chinese New Year |
| 2026/3/4 | Touched terminal, Claude + deployment |
| 2026/3/15 | Created company finance system + OB project management tool |
| 2026/04 | Taught colleagues vibe coding + launched workshops |
Five Steps of Vibe Coding
- Discover Problems: Such as cross-platform reporting causing accounting gaps. At the time, finances were scattered across 7–8 platforms like Simpany, nuieip, Asana, etc.
- Think of Solutions: Decided to integrate multiple systems or build an all-in-one platform, with the core goal of "no missing accounts."
- Map Current Workflows: Analyze work content and pain points of business, finance, and operations roles in the financial process.
- Design Architecture: Plan relationships between modules like project costs, contracts, and sales.
- Build and Deploy: Use rapid deployment with continuous optimization; launch the minimum viable product first and iterate daily based on user feedback.
My Self-Learning Philosophy
Metalearning: Understanding how to learn something new. When learning something completely new, I believe in "metalearning"—first understanding the communication language and knowledge of that field, allowing me to quickly integrate into the field's way of thinking, then take action, try and learn. Through repetitive practice, I expand and deepen my knowledge boundaries.
"From self-learning to teaching is truly learning": To understand the essence of something, beyond learning it yourself, you must also be able to share it with others. That's true mastery.
Consult experts and continuously break through learning and sharing circles: "Seeking answers within incorrect understanding only yields incorrect solutions." Learning something means consulting experts to quickly break through blind spots and expand cognitive boundaries.
Learning vibe coding to deployment has truly brought tremendous change to my surroundings, and collaborating as a team is invaluable. We're not just learning new technology; we're shifting mindsets and strengthening our self-learning ability. Our previous communication was execution-focused, but now it's about "how can we be more efficient," "how can we find solutions," "how can we reflect on workflows," "how can we preserve knowledge"—building an organization with a growth mindset is truly the most fortunate thing!
